Where is the image frame timestamp set in the VI module driver?

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Orin
,
1

Hi NV，

JP5.1.2，AGX Orin
Where is the image frame timestamp set in the VI module driver?

I found the following：
kernel/nvidia/drivers/media/platform/tegra/camera/vi/vi2_fops.c
475 #if KERNEL_VERSION(5, 4, 0) > LINUX_VERSION_CODE
476 getrawmonotonic(&ts);
477 else
478 ktime_get_ts64(&ts);
479 endif

and ：
kernel/nvidia/drivers/media/platform/tegra/camera/vi/vi5_fops.c
558 #if LINUX_VERSION_CODE < KERNEL_VERSION(5, 4, 0)
559 ts = ns_to_timespec((s64)descr->status.sof_timestamp);
560 else
561 ts = ns_to_timespec64((s64)descr->status.sof_timestamp);
562 endif

I want to know where the timestamp in the image frame structure is set? Is this timestamp UTC time?

Thanks

3

hello wangxiaojunfuture,

there’re different version of VI drivers, you should check vi5 as you’re working with Orin series. and… JP-5.1.2 is using kernel-5.10.

hence,
it’s ts = ns_to_timespec64((s64)descr->status.sof_timestamp); to report SOF timestamp to vb2 buffer.
this is TSC HW timestamp, you may see-also Argus::Ext::ISensorTimestampTsc::getSensorSofTimestampTsc for reference.

4

Hi Jerry,

I want to calculate the corresponding realtime (wall time) based on the current capture sof_timestamp(TSC)

  1. Get the TSC domain timestamp from VB2 buffer,
  2. Calculate the realtime boot time
  3. Calculate realtime timestamp of capture

Our sample code:

//1. Get the TSC domain timestamp from VB2 buffer,
struct v4l2_buffer bufferinfo;

bufferinfo.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
bufferinfo.memory = V4L2_MEMORY_DMABUF;

ioctl(camera.fd_, VIDIOC_DQBUF, &bufferinfo)
TSC_time = (bufferinfo.timestamp.tv_sec *1000 + bufferinfo.timestamp.tv_usec / 1000)

//2. Calculate the realtime boot time
clock_gettime(CLOCK_MONOTONIC, &ts);
gettimeofday(&cur_time, NULL);
BootTime = (cur_time.tv_sec * 1000 + cur_time.tv_usec / 1000) - (ts.tv_sec*1000 + ts.tv_nsec/1000000); //power on time

//3. Calculate realtime timestamp of capture
//offset_ns :cat /sys/devices/system/clocksource/clocksource0/offset_ns
RealTimeStamp = BootTime + TSC_time - offset_ns

Are there any logical errors in the above code?

Thanks！

5

hello wangxiaojunfuture,

there’s offset_ns to record the delta from TSC to start of MONOTONIC_RAW.
you may see-also Topic 159220 for an example to use offset_ns compensate the capture timestamp.

6

Hi Jerry，

Do TSC and MONOTONIC_RAW use the same hardware clock source?
If the system runs for 720 hours, will the difference between them become larger and larger?

Thanks a lot！

7

hello wangxiaojunfuture,

to be honest, I’ve never test it for such long period (i.e. 30-days)
as you may know, even the RTC accuracy is -11 ~ +20 seconds/day.

8

Hi Jerry，

Please help ask your company’s R&D personnel about the accuracy or error rules of TSC. Our products involve the fusion of each frame of image and radar data, and they work for a long time, so they must have very accurate timestamps.

Thanks a lot！

9

Please confirm：
Do TSC and MONOTONIC_RAW use the same hardware clock source?

10

hello wangxiaojunfuture,

they’re different. TSC is taking the timestamp of RCE, it’s a dedicated processor for camera management.
anyways, while EGL stream capture the timestamp when it rendering the frame to display. it’s close to end-of-frame of the sensor signaling, you may have alternative way by using EGLStream::IFrame::getTime
for instance, /usr/src/jetson_multimedia_api/argus/include/EGLStream/Frame.h

class IFrame : public Argus::Interface
{
public:
...
    /** 
     * Returns the timestamp of the frame, in nanoseconds.
     */
    virtual uint64_t getTime() const = 0;
11

Hi Herry,

from Topic 159220:
clock_gettime(MONOTONIC_RAW) = cycle_ns(TSC) - offset_ns

Why is the time format not MONOTONIC?

12

because the time has a fixed offset from CLOCK_MONOTONIC_RAW.

14

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.